# Rumi Alexander Elías Calles Director of Research Engineering Director of Research Engineering @ micro1. BSc Data Science; MA Humanities. Classical languages, philosophy of mind, and AI. - Email: rumi.calles@gmail.com - Site: https://rumicalles.com - [LinkedIn](https://www.linkedin.com/in/rumi-calles/) - [GitHub](https://github.com/RumiAllbert) ## About Director of Research Engineering @ micro1. BSc Data Science; MA Humanities. Classical languages, philosophy of mind, and AI. [About](https://rumicalles.com/about) ## Languages - English - Spanish - Chinese - Italian - Latin - Ancient Greek ## Experience ### Director of Research Engineering @ micro1 2025–Present · Remote RL environments and post-training. - Ship RL environments for production post-training: task specs, rewards, eval loops - Own post-training: data, reward models, policy iteration - Set standards for environment quality, training stability, and eval ### Founder & Researcher @ Aeterna Institute 2025–Present · New York, NY Classical liberal arts. Trivium, Quadrivium, Socratic method. - Founded the institute. Curriculum on the Seven Liberal Arts - Coursework from primary texts and dialectic - Small community of fellows ### AI Researcher @ Wolfram Institute 2024–Present · New York, NY Minds, machines, intelligence. Conceptual spaces, AI philosophy, consciousness. - Machine ethics, bias, and explainability - Abliterator: activation engineering for LLM personality - Experiments in Mathematica and Python ### AI Engineer @ Stealth 2024–2025 · New York, NY Production LLM systems for NLU and recommendation eval. - LLM pipelines that lifted NLU quality ~40% across domains - Eval platform for recommendation models. Accuracy up ~30% - Model implementation through production ### Adjunct Professor @ Fei Tian College 2025–2026 · Middletown, NY Ethics of computing. AI governance. - Teach ethics of computing, AI governance, and privacy - Primary frameworks with data-ethics cases - Course design and instruction ### Data Scientist @ The Epoch Times 2022–2024 · New York, NY Predictive modeling, pipelines, and NLP for a publisher. - Churn models that cut subscriber attrition ~20% - Python/Hive pipelines. Ops throughput up ~25% - BERT and clustering over unstructured content - Automated reporting for executives - Projects from framing through deployment - Measurement and modeling with product and editorial ### CTO & Executive Board Member @ Students for Falun Gong 2019–Present · Remote Technical lead for an international nonprofit. - Multilingual site and servers - International conferences and logistics - Advocacy at the International Religious Freedom Summit - Digital infrastructure - Work with international teams ### Data Analytics Research Assistant @ UD Social Neuroscience Lab 2021–2022 · Delaware Python stats for social neuroscience experiments. - PsychoPy tooling that cut the lab's data pipeline ~30% ### Data Scientist @ Litespeed Technologies 2021–2021 · Middletown, NY Data mining and reporting for executives. - Mining scripts that cut processing time about in half - Analysis across company data streams ### Data Analytics Alumni @ NASA 2019–2020 · Middletown, NY Time series and geospatial analysis. - USGS geospatial maps under NASA, UN, and USGS mentorship ### Full-Stack Developer & Assistant Creative Director @ The Werking Title Gallery 2020–2020 · San Diego, CA Web platform and client sites for a gallery. - Shipped the gallery site and client sites ## Publications ### De Suspensione: Inflection, Composition, and the Long Drift Outward Rumi Alexander Elías Calles · 2026 · ResearchGate On suspensio, the inner act of holding thought while form catches up. Latin inflection, digital memory, and generative AI. Three historical migrations — the Latin-to-Romance simplification, the digital externalization of memory, and the offloading of sentence production to generative AI — converge on a single mental practice: suspensio, the inner act of composition once trained by the Trivium and by case-marking grammars that obliged the speaker to track relation internally. Drawing on classical-language impact studies, Romance morphosyntactic typology, neuro- and psycholinguistic work on grammar and thought, and recent empirical work on cognitive offloading and large language models, the paper rejects the strong Whorfian thesis that simplification produced cultural shallowness, defending instead the narrower claim that form, content, and production now migrate outward together — and that compositional practice can thin under AI assistance. [https://www.researchgate.net/publication/405343127_De_Suspensione_Inflection_Composition_and_the_Long_Drift_Outward](https://www.researchgate.net/publication/405343127_De_Suspensione_Inflection_Composition_and_the_Long_Drift_Outward) ### Uncovering Candidate Potential with Multi-Modal AI Assessments Rumi Alexander Elías Calles et al. · 2025 · ResearchGate AI Match Score beats résumé screening by 16% AUC and finds 2× more qualified candidates. Our AI Match Score combines resume analysis, structured interviews, and role-specific evaluations to outperform traditional screening by 16% (AUC: 0.742 vs. 0.641). It surfaces nearly twice as many qualified candidates, capturing unique potential signals beyond what resumes reveal. [https://www.researchgate.net/publication/395809850_Uncovering_Candidate_Potential_with_Multi-Modal_AI_Assessments](https://www.researchgate.net/publication/395809850_Uncovering_Candidate_Potential_with_Multi-Modal_AI_Assessments) ### From Plato to Pixels: Imagination, AI, and the Whole Rumi Alexander Elías Calles et al. · 2025 · ResearchGate Imagination from Plato's Forms to generative AI. What machines still cannot make. This philosophical exploration traces imagination's evolution from Plato's Forms through medieval theology to modern fragmentation, examining whether AI can truly simulate human creativity. It argues that imagination remains humanity's essential faculty for synthesizing meaning from chaos, even as machines attempt to replicate creation. [https://www.researchgate.net/publication/395101656_From_Plato_to_Pixels_Imagination_AI_and_the_Whole](https://www.researchgate.net/publication/395101656_From_Plato_to_Pixels_Imagination_AI_and_the_Whole) ### Evaluating Speech-to-Text x LLM x Text-to-Speech Combinations for AI Interview Systems Rumi Alexander Elías Calles et al. · 2025 · arXiv Voice-stack comparison across 300k+ AI-conducted interviews. We present a large-scale empirical comparison of speech-to-text, LLM, and text-to-speech combinations using data from over 300,000 AI-conducted job interviews. Our analysis reveals that Google's STT with GPT-4.1 and Cartesia's TTS outperforms alternatives in both objective quality metrics and user satisfaction scores. [https://arxiv.org/abs/2507.16835](https://arxiv.org/abs/2507.16835) ### PhiloBERTA: A Transformer-Based Cross-Lingual Analysis of Greek and Latin Lexicons Rumi Alexander Elías Calles, Makai L. Allbert · 2025 · arXiv Cross-lingual transformer on ancient Greek and Latin lexicons. A cross-lingual transformer model measuring semantic relationships between ancient Greek and Latin lexicons, revealing significant alignments in philosophical concepts. [https://arxiv.org/abs/2503.05265](https://arxiv.org/abs/2503.05265) ### Identifying and Manipulating Personality Traits in LLMs Through Activation Engineering Rumi Alexander Elías Calles, et al. · 2024 · arXiv Abliterator. Steering LLM personality through activation engineering. We introduce Abliterator, a framework for studying and manipulating personality traits in large language models (LLMs) through activation engineering. [https://arxiv.org/abs/2412.10427](https://arxiv.org/abs/2412.10427) ## Skills ### Focus AI Research, Machine Learning, NLP & LLMs, Data Science, Research Engineering ### AI & ML PyTorch, Transformers, LangGraph, MCP, OpenEnv, Harbor, vLLM, LiteLLM, TransformerLens, Verifiers, Unsloth, wandb ### Languages & Data Python, TypeScript, Go, R, Mathematica, Pandas, Polars, NumPy, Plotly, Streamlit ### Infrastructure & Tooling Docker, Kubernetes, FastAPI, Ollama, Git, GitHub, AWS, Google Cloud, Spark, PostgreSQL, Redis ## Education ### Ralston College Master of Arts · Humanities · 2023–2024 - Full-tuition scholarship - Greek Language Residency - Study in Greece and Turkey ### Fei Tian College Middletown Bachelor of Science · Data Science · 2019–2023 - First place, hackathon - Academic excellence award ### Academia Vivarium Novum Immersive Program · Classical Studies and Philosophy · 2024–2025 - Lived in Latin and Ancient Greek - Primary texts, philosophy, and rhetoric in the original languages ### Wolfram Summer School Certificate · Artificial Intelligence & Philosophy · 2024–2024 - Research in AI and large language models - Wolfram Language experiments ### National Taiwan Normal University Language Studies · Mandarin Chinese · 2023–2023 - Huayu Scholarship recipient ### Center for Hellenic Studies, Harvard University Visiting Greek Scholar · Greek Studies · 2023–2023 - Full-time Greek residency - Seminars with visiting Hellenists ### Certifications - Ethics of Artificial Intelligence (Politecnico di Milano, 2026) - Philosophy of Science (University of Pennsylvania, 2026) - Applied AI Product Management (IBM, 2025) - Deep Learning (Lightning AI, 2025) - CrewAI Multi-Agent Systems (CrewAI, 2025) - MCP Automation (Hugging Face, 2025) - AI Agents (Hugging Face, 2025) - Neural Networks & Deep Learning (DeepLearning.AI, 2022) - Database Structures & Management (Meta, 2022) - Version Control (Meta, 2022) - Advanced Google Analytics (Google, 2022) - Credit Risk Modeling in R (DataCamp, 2020) - R for Finance (DataCamp, 2020) - Human Rights Defense (Amnesty International, 2019) ## Projects ### Interlinea Reading companion for Latin and Ancient Greek. Every word glossed, parsed, and linked to its sentence. - Homer, Plato, Virgil and Caesar in the original, word by word - Case-colored grammar and syntax links that rejoin scattered Latin word order - Glaux, a step-by-step tutor that translates only when asked [https://interlinea.aeterna-institute.org/](https://interlinea.aeterna-institute.org/) ### Vertor Translation workspace. Compose, translate, and refine literary text across models. - Sentence-by-sentence streaming across Gemini, Claude, and GPT with paragraph-aware structure - Word-level refinement: literal, natural, and lyrical alternatives - Compose–translate–refine workflow on Next.js [https://vertor.vercel.app/](https://vertor.vercel.app/) ### Etymon.ai Etymology engine. Morphological roots, with emphasis on classical languages. - Morphological analysis for classical languages and semantic change - React Flow graphs of linguistic relationships - Gemini-backed morphological breakdown [https://etymon-ai.com/](https://etymon-ai.com/) ### Memento Vitae Life in remaining weeks. Time allocation and mortality accounting. - Life grid of ~4,000 weeks with current-week tracking - Breakdown of how sleep, work, and habits consume remaining time - Concrete counts: visits left, books, summers remaining [https://mementovitae.vercel.app/](https://mementovitae.vercel.app/) ### Loqui TTS Local TTS on Apple Silicon. Voice cloning, multilingual, no cloud. - Voice cloning and 10-language support via Qwen3-TTS - Metal GPU inference with live system monitoring - React + FastAPI with in-browser model downloads [https://github.com/RumiAllbert/loqui-tts](https://github.com/RumiAllbert/loqui-tts) ### Latinium.ai Latin tutor with live grammar analysis and interactive annotation. - NLP grammar engine for case, tense, and syntax - Hover annotations with model-backed tutoring - Translation that preserves literary register [https://latinium.rumiallbert.com/](https://latinium.rumiallbert.com/) ### Gentleman's Codex Guide to men's style. Cut, color, rules. - Four-chapter curriculum with quizzes and style archetype assessment - Outfit builder with compatibility scoring - Color wheel and curated palettes [https://gentlemans-codex-1031601420066.us-central1.run.app/](https://gentlemans-codex-1031601420066.us-central1.run.app/) ### ColorLab.ai Color analysis from selfies. Seasonal palettes and style guidance. - Selfie analysis on Next.js - Model-backed seasonal palette recommendations - LangChain workflows for style guidance [https://color-lab-ai.vercel.app/](https://color-lab-ai.vercel.app/) ### LLM Abliterator Toolkit for steering LLM personality through activation engineering. - Identify and edit personality-related activation vectors - Personality steering with low compute overhead - Interpretability tests with ethical analysis [https://community.wolfram.com/groups/-/m/t/3210605](https://community.wolfram.com/groups/-/m/t/3210605) ### Ancient Greek Resources Classical Greek. Texts, grammar, lexicography. - Annotated texts, grammar references, and lexicographical tools - Spaced repetition for morphology - Content standards aimed at classroom use [https://www.ancientgreek.rumiallbert.com/](https://www.ancientgreek.rumiallbert.com/) ## Contact - [Contact](https://rumicalles.com/contact) - [Privacy](https://rumicalles.com/privacy) - [llms.txt](https://rumicalles.com/llms.txt) - [Sitemap](https://rumicalles.com/sitemap.md)